My intention in running Annual Water Yield was to find the contribution of agricultural areas to water yield. I had some doubts reading the guide in this specific assumption:
“Outputs in the per_pixel folder can be useful for intermediate calculations but should NOT be interpreted at the pixel level, as model assumptions are based on processes understood at the subwatershed scale”
Is it possible to consider contributions from each LULC class to water yield volume inside the same subwatershed?
Thanks for the question. @swolny and others can correct me here, but I think a good approach to considering contributions from each LULC class to the water yield volume in a given watershed is to create scenarios.
If we were interested in knowing how agriculture vs development vs conservation (forests perhaps) impacted water yield volume, we might run the model first with agriculture LULC values and then change those values or a subset of those values to developed. Running the model again we could compare the difference in output. We could then do the same thing for our conservation scenario. I think that these kinds of watershed / sub-watershed comparisons are what the model was designed for; getting a sense of how different landscape practices might effect annual water yield outcomes.
Unlike our other water models (Seasonal Water Yield, SDR, NDR), Annual Water Yield does not do any routing of surface water. Because of this, I suppose looking at the pixel output would tell you the straight up difference between LULC types since there are no interactions between them. It would be the same as using the equations in the Users Guide to compare. HOWEVER, I believe the equations used were designed for the watershed / sub-watershed level, so use those values relatively and not as absolute truths.
As @dcdenu4 said, we do generally use scenarios when considering changes between land cover types. If your analysis does not involve scenarios, I do feel like it’s reasonable to aggregate the water yield per-pixel values over different land cover types (like agriculture), taking the sum and/or mean and comparing them between land cover types.
As Doug also said, the underlying model equations are designed for the sub-watershed scale. So if you look at the per-pixel maps, we would not feel comfortable reporting absolute values at the pixel scale, but we do feel more comfortable looking at the overall pattern of high and low water yield, which can be compared with other studies if available.
I have the same objective in one of my papers using the pixel level values of AWY model, the overall contribution of the LULC, and possible change if a certain scenario is implemented. Just want to share my experience, assuming that all the metric adjustments were done, you should be able to compute for the total contribution per land cover by aggregating the per pixel result. Then the total per land cover can be divided to the total area of that land cover to get an average per pixel contribution. The average per pixel could at least provide good information on the implications of the AWY per pixel tying to ecosystem service changes. The average per pixel could also provide good insights if scenarios land cover changes are implemented. Here is a link to our paper for your reference. https://doi.org/10.3390/land10010021
Hi Stacie and Doug,
Please correct me if I am wrong.
Thank you for your guidance on this.